2. visual perception: optimizing information visualization regarding the human visual ... ·...
TRANSCRIPT
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
2. Visual Perception:Optimizing Information Visualization regarding the human visual system
Vorlesung „Informationsvisualisierung”Prof. Dr. Andreas Butz, WS 2011/12Konzept und Basis für Folien: Thorsten Büring
1
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Outline• Perception Definition & Context• Preattentive processing• Gestalt Laws• Change Blindness• Data encoding – glyphs• Data encoding – color• Characteristics of Visual Properties
2
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Perceptual Processing• Design visual information to be efficiently
perceivable – quick, unambiguous• Need to understand how human visual perception
and information processing works• Perception science related to:
–Physiology: study the physical, biochemical and information processing functions of living organisms
–Cognitive psychology: studying internal mental processes• how do people learn, understand, solve problems with regard
to sensory information?
3
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Model of Perceptual Processing• Numerous other models exist• Simplified 3-stage model: many subsystems
involved in human vision• Stage 1 – rapid parallel processing to extract low-
level properties of a visual scene–Detection of shape, spatial attributes, orientation, color,
texture, movement–Billions of Neurons work in parallel, extracting information
simultaneously –Occurs automatically, independent of (cognitive) focus– Information is transitory (though briefly held in a short-
lived visual buffer)–Often called “preattentive” processing
4
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Model of Perceptual Processing
Image taken from Ware 20015
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Model of Perceptual Processing• Stage 2 – pull out structures via pattern perception
– Visual field is divided in simple patterns: e.g. continuous contours, regions of the same color / texture
– Object recognition– Slower serial processing
• Stage 3 – sequential goal-directed processing– Information is further reduced to a few objects held in visual
working memory– Used to answer and construct visual queries– Attention-driven - forms the basis for visual thinking– Interfaces to other subsystems:
• Verbal linguistic: connection of words and images• Perception-for-action: motor system to control muscle movement
6
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Example
• Route between the two letters?• Stage 1: automatic parallel extraction
of colors, shapes, position etc.• Stage 2:
–Pattern finding of black contours (lines) between two symbols (letters)
• Stage 3: –Few objects are held in working memory at a time– Identify path sequentially (formulate new visual query)
• In this lecture we will focus on aspects related to stage 1 & 2 of the model
7
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Outline• Perception Definition & Context• Preattentive processing• Gestalt Laws• Change Blindness• Data encoding – glyphs• Data encoding – color• Characteristics of Visual Properties
8
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Preattentive Processing• A limited set of basic visual properties are
processed preattentively• Information that “pops out”• Parallel processing by the low-level visual system
(Stage 1 in the model) • Occurs prior to conscious attention• Important for designing effective visualizations
–What features can be perceived rapidly?–Which properties are good discriminators?–What can mislead viewers?–How to design information such that it pops out?
9
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Example: Find the 3s
142416496357598475921765968474891728482285958819829450968504850695847612124044074674898985171495969124567659608020860608365416496457590643980479248576960781285960799918712845268101495969124567781874241649645757659608149596912456701285960799164964575127879918712845298496912223591649645759588198250963576596080596
10
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Example: Find the 3s
142416496357598475921765968474891728482285958819829450968504850695847612124044074674898985171495969124567659608020860608365416496457590643980479248576960781285960799918712845268101495969124567781874241649645757659608149596912456701285960799164964575127879918712845298496912223591649645759588198250963576596080596
11
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Preattentive Processing• How to find out if a visual attribute is preattentive?• Measure response time for tasks
– Detection of a target among distractors – Is the target present?
– Boundary detection – Do items form two groups?– Counting – How many targets are there?
• Detection of targets on a large multi-element display – < 200 to 250 ms are considered preattentive – Eye movement takes at least 200 ms to initiate
• Example: is there a red target present in the images?
12
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Color• Is there a red circle present in the image?
Images taken from http://www.csc.ncsu.edu/faculty/healey/PP/index.html
13
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Color• Is there a red circle present in the image?
Color is preattentively processed!
14
Images taken from http://www.csc.ncsu.edu/faculty/healey/PP/index.html
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Shape• Is there a red circle present in the image?
15
Images taken from http://www.csc.ncsu.edu/faculty/healey/PP/index.html
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Shape• Is there a red circle present in the image?
Shape is preattentively processed!
16
Images taken from http://www.csc.ncsu.edu/faculty/healey/PP/index.html
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Color & Shape• Is there a red circle present in the image?
17
Images taken from http://www.csc.ncsu.edu/faculty/healey/PP/index.html
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Color & Shape• Is there a red circle present in the image?
Conjunction of 2 properties is usually not preattentive!
18
Images taken from http://www.csc.ncsu.edu/faculty/healey/PP/index.html
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Boundary Detection• Do items form a boundary? If yes, based on which
attribute(s)?
19
Images taken from http://www.csc.ncsu.edu/faculty/healey/PP/index.html
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Boundary Detection• Do items form a boundary? If yes, based on which
attribute(s)?
Conjunction search: grouping by hue and shape
Preattentive: grouping by hue
20
Images taken from http://www.csc.ncsu.edu/faculty/healey/PP/index.html
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Common Preattentive Properties• Color
–Hue–Intensity–Motion–Flicker–Direction of motion–Spatial Position–2D position–Stereoscopic depth–Convexity / Concavity
Images taken from http://www.csc.ncsu.edu/faculty/healey/PP/index.html
21
• Form–Line orientation–Line length–Line width–Size–Curvature–Shape–Spatial grouping
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Conjunction Search
• A target with a unique visual property (e.g., shape OR color) “pops out”
• Conjunction target is made up of non-unique features–Requires a time-consuming serial
search, e.g.– For every red colored item: is it a circle? – For every cricular item: is it red?
22
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Outline• Perception Definition & Context• Preattentive processing• Gestalt Laws• Change Blindness• Data encoding – glyphs• Data encoding – color• Characteristics of Visual Properties
23
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Cognition and Gestalt Laws• See „Digitale Medien“, Chapter 3a (typography)• Recap: step 2 of the visual information processing model –
pattern and object recognition using the raw data collected in step 1
• Based on which visual properties do we structure the data?• Gestalt school of psychology founded in 1912 formulated
Gestalt laws• “The whole is greater than the sum of parts” (Koffka 1935)• Laws still useful today, but not the neural mechanisms
proposed• Perception: An introduction to the Gestalt-theorie (Kurt
Koffka, 1922): http://psychclassics.yorku.ca/Koffka/Perception/perception.htm
24
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Example of Gestalt perception
25
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
What do you see?• Can you find the dog?• Dalmatinian exploring a leave covered forest floor• Once you have found it, try to think of the picture
as a simple pattern of black and white again• Does it work?• Mind tries to detect anything meaningful by
identifying patterns• Different tools are tried sequentially• Perceptual organization is a powerful mechanism
26
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
GL: Grouping by Spatial Proximity• Columns or rows?• Small difference in spacing causes change in perception• Use proximity to emphasize between display items• To which group (top / bottom) does the x dot belong?
Spacing is equal for both groups!• Spatial concentration principle: we group regions of similar
element density (Slocum1983)
x
27
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
GL: Similarity • Rows or columns?• Similar elements tend to be grouped together
28
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
GL: Connectedness• Palmer & Rock 1994• Potentially more powerful organizing principle than
proximity, color, size, shape
proximity color size shape Example: node-link diagram
29
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
GL: Continuity• Example circuit design – understanding how
components are connected
30
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
GL: Symmetry
• Example of how symmetry detection may be exploited for visual data mining
• Support the search for similar patterns in time-series plots (measurements of deep ocean drilling cores)
Image taken from Ware 2001
31
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
GL: Area• Smaller components of a pattern tend to be
perceived as an object • White propeller and black propeller
32
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
GL: Figure & Ground• Figure: something object-like that is perceived
being in the foreground• Ground: whatever lies behind the figure• Fundamental perceptual act of identifying objects• All Gestalt laws contribute, e.g., closed contour,
symmetry, area• Equally balanced cues for figure and ground can
result in bistable perception
33
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
GL: Common Fate• Objects moving in the same direction are
perceived as an entity• Example taken from: http://tepserver.ucsd.edu/
~jlevin/gp/time-example-common-fate/
34
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Outline• Perception Definition & Context• Preattentive processing• Gestalt Laws• Change + Inattentional Blindness• Data encoding – glyphs• Data encoding – color• Characteristics of Visual Properties
35
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Change Blindness (CB)• Example: old style aircraft altimeter
– Thinnest hand indicates number of tens of thousands of feet
– Next larger hand number of thousands of feet– Quick glance after interruption results in
misinterpretation if the change in the display is not noticed
– Difference of ten thousand feet
• Phenomenon: inability to detect changes in visual scenes– mid-eye movement– mid-blink– Flicker (short blanking of screen)– Gradual change
36
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Change Blindness (CB)• Participants of a study were found unable to detect a
change from one person to another in midconversation (Simson & Levin 1998)
• Sample principle: insensitivity to changes of objects in movie scenes interrupted by a cut (Levin & Simons 1997)
• Various examples: http://viscog.beckman.uiuc.edu/djs_lab/demos.html
• Problem related to the short-lived visual buffer and the very limited capacity of our visual working memory
• Need to emphasize changes• In some applications changes may be distracting, e.g.
ambient information visualization -> utilize CB37
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
CB: Flicker Example 1
http://www.csc.ncsu.edu/faculty/healey/PP/index.html
38
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
CB: Flicker Example 2
http://www.csc.ncsu.edu/faculty/healey/PP/index.html
39
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
CB: Gradual Change Example
40
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Recognition is primed by expectation• http://www.youtube.com/watch?
v=kjtSfTCrMm4&feature=related
41
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Inattentional Blindness• Basketball passes: http://www.youtube.com/
watch?v=Ahg6qcgoay4&feature=fvw• Color changing card trick: http://www.youtube.com/
watch?v=hDLLf_WCyZ4&hl=de
42
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Perception Summary
• It is not enough to simply show something.• We need to pay attention when and how it
is shown.• Otherwise people might miss it or take
forever to find it.
• Apply your knowledge about perception to check whether your visualizations are effective!
45
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Outline• Perception Definition & Context• Preattentive processing• Gestalt Laws• Change Blindness• Data encoding – glyphs• Data encoding – color• Characteristics of Visual Properties
46
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Encoding Data with Glyphs• Glyph: graphical object designed to convey multiple data values• Multidimensional discrete data, e.g. a collection of cars with
several attributes each, e.g. horsepower, weight, acceleration etc.• What visual properties can be mapped to the attributes?• FilmFinder example
– Color to film genre– X-position to year of release– Y-position to popularity
• Additional properties– Lightness– Shape– Orientation– Texture– Motion– Blinking
FilmFinder (www.cs.umd.edu/) 47
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Encoding Data with Glyphs• Limitations of low-level graphical attributes for glyph design• Easily resolvable steps of a visual property
– 12 colors (for preattentive processing only 8 colors)– About 4 orientation steps– At most 4 size steps– Binary blink coding (on / off)– Texture unknown– Shape unkown
• Mixing visual properties– Properties are not independent from each other, e.g. blink coding
interferes with motion coding– Conjunctions are usually non-preattentive– Some dimensions are integral– Best to restrict the mapping to color, shape, spatial position (and motion)
48
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Outline• Perception Definition & Context• Preattentive processing• Gestalt Laws• Change Blindness• Data encoding – glyphs• Data encoding – color• Characteristics of Visual Properties
49
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Color Vision & Model• Human color vision
– Sensory response to electromagnetic radiation in the spectrum 0.4 – 0.7 micrometers
– Based on three dimensions (three different types of color receptors in human retina)
• Powerful encoding potential: compared to gray scales the number of just noticeable differences is much higher
• About 8% of the male and 1% of the female population are color-blind
• Color Model HSV (aka HSB)– Hue - blue, green, etc. (X axis)– Saturation – intensity of color (Y axis)– Value – light/dark (slider)
HSV cone representation
HSV 2D color picker
50
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Color Scales• Definition: pictorial representation of a set of
distinct categorical or numerical values, where each value is assigned its own color (Levkovitz 1996)
• Desired properties of perception–Preserve the order of the data values, if any–Uniform distance between adjacent values (i.e. equally
spaced numerical steps are perceived as equally spaced perceptual steps)
–No artificial boundaries that do not exist in the data (i.e. continuously present continuous values)
51
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Color Rules I• Always ensure a reasonable luminance contrast
between foreground and background color – chromatic variation may not be enough!
• Black and white borders around colored symbols can reduce contrast effects
• Canonical colors (close to an ideal) are easier to remember
• Only a small set of basic colors should be used for nominal (distinct) labeling– At most 12 colors: red, green, yellow, blue, black, white,
pink, cyan, gray, orange, brown, purple– The first four colors are “hard-wired” into the human brain –
should be used with priority
52
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Grayscale• Usually not considered a color scale, but very common• Provides simple and natural sense of order • Disadvantages
– Limited number of just-noticeable-differences (JNDs) about 60 to 90– Contrast effects can significantly reduce accuracy– Luminance channel is fundamental to much of perception –
grayscale encoding may be considered “a waste of perceptual resources” (Ware, 2000)
53
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Rainbow for Ordering Data?• Most common: rainbow scale for ordinal and
quantitative (spectral colors)–Continuous spectrum –Common arbitrary division in 8 or less named colors (red,
orange, yellow, green, cyan, blue, indigo, violet)• Problems with rainbow scale
–Can you order the color blocks from low to high?–Yellow (in the middle of the scale) may draw too much
attention, when users are seeking for extreme values–Perception of non-existing boundaries
John Stasko54
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Recommended Color Scales• Ordinal data
–Low saturation to high saturation (single hue) - also very limited JNDs
–Dark to light (single hue)–Red to green, yellow to blue, red to blue
• Ratio (hardly feasible) / diverging data–Neutral value (e.g. white) to represent zero– Increases in saturation toward distinct colors for positive
and negative values (double-ended multiple hue)• Look up www.colorbrewer2.org for inspiration
55
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
colorbrewer2.org
56
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Redundant Color Scales• Use multiple color properties to redundantly represent data• Visual reinforcement of steps• Overcome visual deficiencies• Redundant model components: data values are mapped to both hue and
brightness• Heated-object scale
– Going from black to white passing through orange and yellow– Monotonic increase in brightness provides more natural ordering than rainbow scale
• Linearized optimal color scale– Scale maximizing the number of JNDs while preserving a (more or less) natural order
Silva et al. 200757
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Color Scale• US presidential elections - Bush & RNC's
campaign funding
$7,000,000
$20
http://fundrace.huffingtonpost.com/moneymap.php?cand=Bush&zoom=County
58
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Color Scale• Vote distribution of 2004 US presidential election -
the darker the color, the more of a landslide it was for the winning party
Republican
Democrat
Neutral
http://fundrace.huffingtonpost.com/moneymap.php
59
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Color Scale
Sheelagh Carpendale
60
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Color Scale
Sheelagh Carpendale
61
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Color Rules II• For larger areas on a white background use low-saturation
light colors• Small color-coded objects should be given high saturation• Use red and green in the center of the field of view (edges
of retina not sensitive for these)• Use black, white, yellow in periphery• Use color for grouping and search• Color Blindness Simulator: http://www.etre.com/tools/
colourblindsimulator/ • Generation of color families
– Use canonical colors– Family members should differ by saturation – Better: saturation and lightness
62
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Bivariate Color Coding• Recap: color is three-dimensional• Two data dimensions may be mapped to different color
dimensions (e.g. hue and saturation, hue and lightness)• Problem: bivariate color coding has been found
notoriously difficult to read (Wainer & Francolini, 1980)• The same applies to multidimensional color coding
– E.g. amount of red, amount of green, amount of blue for coding colored dots in scatterplot (Ware & Beatty 1988)
– Clusters could be easily identified by the participants of a user test– Precise decoding of the color components difficult
63
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Example: Bivariate Color Coding
Image by Cynthia Brewer, http://www.psu.edu/ur/2003/MapColorphotos.html
64
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Outline• Perception Definition & Context• Preattentive processing• Gestalt Laws• Change Blindness• Data encoding – glyphs• Data encoding – color• Characteristics of Visual Properties
65
LMU München – Medieninformatik – Andreas Butz – Informationsvisualisierung – WS2011/12 Folie
Characteristics of Visual Properties
• Some properties possess intrinsic meaning– Density with Grayscale: the darker the
more– Size / Length / Area: the larger the more– Position: depending on culture, in Europe
the leftmost / topmost are first– Color: depending on culture, e.g. white
associated with death in Japan
• Accuracy of representations for quantitative measures (empirically verified by Cleveland & McGill, 1985)
(Mackinlay, 1988)
Attribute Quantitative Qualitative
Line length x
2D position x
Orientation x
Line width x
Size x
Shape x
Curvature x
Enclosure x
Hue x
Intensity x
(Few, 2004)66